OpenAI is introducing a new ad format in ChatGPT that will show display ads to US users while the chatbot generates an image. The ads appear as product carousels on the image generation loading screen, expanding advertising within ChatGPT.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
OpenAI will test display ads on the image-generation loading screen in US ChatGPT, while rolling out conversion postbacks, attribution, geographic experiments, and brand safety tools. The real story here is not the ad slot itself, but the advertising infrastructure taking shape behind it—and a set of revenue figures backed only by OpenAI itself.
According to THE DECODER, US ChatGPT users will soon see display ads while the chatbot generates images. These ads appear below the image ChatGPT is generating, in the form of a carousel or a row of product images, are labeled as ads, and are kept separate from the generated image.
The ads are designed to show product use cases or related experiences. This is worth noting: it is not about hard-pasting product images next to the result, but about trying to embed into the wait time of "being generated." OpenAI will launch the test with its first advertisers later this month.
Putting ads into the loading screen means the ad slot no longer attaches only to conversation content, but to the user's operational rhythm. This is the core question this article pursues: when ads begin to occupy the blank moments in the product flow, what exactly is OpenAI selling—attention, or a whole set of performance systems that advertisers can verify?
In sync with the new display format, OpenAI is expanding its measurement tools. Advertisers can send conversion data to ChatGPT Ads through Hightouch, Tealium, and LiveRamp, and attribution providers such as AppsFlyer, Adjust, and Branch have also joined the system.
On performance measurement, OpenAI is testing geographic-based experiments through Haus, Measured, and WorkMagic to measure the actual impact of ads. On brand safety, OpenAI is running pilot projects with DoubleVerify and Integral Ad Science to evaluate the ad environment without accessing real user conversations. In addition, some advertisers can set negative keywords to avoid their ads appearing in inappropriate conversations.
Looking at these moves together, it can be inferred (inference): OpenAI is pushing advertising from "can it be run" to "how is it counted, how accurate is the counting, and how do we prevent problems." Conversion postbacks, attribution, geographic experiments, brand safety, and negative keywords each correspond to one of the most common gates in advertisers' purchasing decisions.
But uncertainty must be flagged: whether geographic-based experiments can accurately measure the actual impact of ads remains to be verified; whether the brand safety pilot can effectively evaluate the ad environment without accessing real conversations is likewise uncertain. These two points determine that the above infrastructure is currently closer to "under construction" than "completed."
OpenAI has been running ads in ChatGPT since February, currently covering more than 40 countries. The report says its ad business alone has reached an annualized revenue run rate of $1 billion, and the company's goal is to achieve $100 billion in ad revenue by 2030.
These two figures need to be treated with caution: they are both OpenAI's unilateral claims, lacking third-party verification, so in this article they can only be handled as source_claim, not as confirmed facts. Similarly, according to OpenAI, ChatGPT reaches 1.2 billion people weekly, while earlier this year only about 5% of users paid for subscriptions—these two data points are also provided by OpenAI and have not been independently verified.
Putting the $1 billion annualized run rate side by side with the $100 billion 2030 target, the gap itself is information: it shows that advertising is set as a curve requiring long-term steep growth, not a one-time supplementary revenue. But whether this curve can hold depends on whether the aforementioned measurement and brand safety tools can truly convince advertisers to keep increasing budgets.
The user structure given in the report provides a clue for explanation: ChatGPT reaches 1.2 billion people weekly, but earlier this year only about 5% of users paid for subscriptions. This combination means the vast majority of users do not directly contribute subscription revenue.
From this it can be inferred (inference): given limited subscription conversion rates, advertising becomes one of the main paths to monetize the massive usage volume. And the image-generation loading screen was chosen because there is a period when users are already waiting and their attention has nowhere to go.
However, the boundary of this inference must be made clear: the report does not disclose the actual impact of this ad format on user experience and the image-generation process, nor does it give specific conversion performance for loading-screen ads. Therefore, "the loading screen is an efficient ad slot" is currently only a design assumption, not a verified conclusion.
The boundaries of this move are quite clear: ads are only being rolled out to US ChatGPT users; ads are labeled as ads and separated from generated images; the brand safety pilot evaluates the ad environment without accessing real user conversations; negative keywords are only open to some advertisers; and the test launches only with the first advertisers.
Stacking these constraints together shows that this is a controlled test rather than a full-scale rollout. The single US market, the first advertisers, and negative keywords available only to some advertisers all point to the same judgment: while expanding advertising, OpenAI is still controlling its risk exposure.
The undisclosed information is equally important: neither the specific launch date of the ad test nor the list of first advertisers has been announced. Without these two items, any estimate of scale or pace can only remain at the level of speculation.
All facts in this article come from a report by THE DECODER (article_id 20490). That report is secondhand in nature, and its source is labeled as OpenAI. The ad format, rollout region, measurement and brand safety partners, ad start time, and number of covered countries are content explicitly stated in the report; while the $1 billion annualized ad revenue run rate, the $100 billion 2030 target, the 1.2 billion weekly reach, and the roughly 5% paying user share are all OpenAI's unilateral claims, lacking third-party verification, and are handled in this article as source_claim. Paragraphs marked as inference are inferences based on the above facts and do not represent confirmed conclusions.
The key to this test is not that "ChatGPT has gained another ad slot," but that OpenAI is simultaneously filling in the pieces of conversion postbacks, attribution, geographic experiments, and brand safety—the ad slot is the visible step, while the measurement system is the step that determines whether advertisers will pay over the long term. As for the $1 billion run rate and the $100 billion target, they are still only OpenAI's own claims.
Primary report
Primary source